Meta releases new open-source AI coding assistant Code Llama 70B

Code Llama 70B scored 53 percent in accuracy on the HumanEval benchmark which is better than GPT-3.5’s 48.1 percent and closer to the 67 percent for GPT-4.

FILE PHOTO: Meta AI has released an updated version of its AI code generation model, Code Llama 70B. | Photo Credit: Reuters

Meta AI has released an updated version of its AI code generation model, Code Llama 70B. In a blog posted on Monday, the company called it the “largest and best-performing model in the Code Llama family.” The model can write code in different programming languages like Python, C++, Java and PHP, and is also able to handle more queries than it could in the past while being more accurate.

Code Llama 70B scored 53% in accuracy on the HumanEval benchmark which is better than GPT-3.5’s 48.1% and closer to the 67% for GPT-4.

Meta had released its previous batch of Code Llama tools called Code Llama Python and Code Llama Instruct around specific coding languages, in August last year. Like them, Code Llama 70B was also built on Llama 2 and helps developers to write snippets of code using prompts and debug their own code.

(For top technology news of the day, subscribe to our tech newsletter Today’s Cache)

Code Llama-70B was trained on a massive 1TB dataset of code and code-related data. The model is currently hosted on the Hugging Face code repository where all three versions of are available for free for research purposes. The inference code for Code Llama models is available on GitHub.